Figures

Figure 1. Modern pollen samples used in this study from the Neotoma Paleoecology Database.

Figure 1. Modern pollen samples used in this study from the Neotoma Paleoecology Database.

Figure 2. The Southwest United States (SWUS) as defined here, showing locations of fossil pollen samples used in this study. Background shading indicates mean July temperature for 1961–1990 from PRISM (Daly et al. 2008; PRISM Climate Group 2019; Daly, Smith, and Olson 2015) at 800-m resolution. The solid gray box outlines the area studied by Bocinsky and Kohler (2014); the green dashed box shows the extent of the reconstruction in this study.

Figure 2. The Southwest United States (SWUS) as defined here, showing locations of fossil pollen samples used in this study. Background shading indicates mean July temperature for 1961–1990 from PRISM (Daly et al. 2008; PRISM Climate Group 2019; Daly, Smith, and Olson 2015) at 800-m resolution. The solid gray box outlines the area studied by Bocinsky and Kohler (2014); the green dashed box shows the extent of the reconstruction in this study.

Figure 3. Temperature for the North Atlantic using Richard Alley’s (2000) Greenland ice core data (black line) and global mean surface temperature change (Osman et al. 2021) (using the 50th ensemble percentile) for 22,000 BC to present with the Younger Dryas and the general period of the northward expansion of maize agriculture from Mesoamerica into the SWUS highlighted.

Figure 3. Temperature for the North Atlantic using Richard Alley’s (2000) Greenland ice core data (black line) and global mean surface temperature change (Osman et al. 2021) (using the 50th ensemble percentile) for 22,000 BC to present with the Younger Dryas and the general period of the northward expansion of maize agriculture from Mesoamerica into the SWUS highlighted.

Figure 4. Low-frequency component of the Moberg et al. (2005) multi-proxy reconstruction for Northern Hemisphere for AD 133–1925. PAGES2k reconstructed NH temperatures using different proxies that primarily shows high frequency resolution (PAGES 2k Consortium et al. 2019). Anomalies for both series are calculated in comparison to the 1961–1990 average.

Figure 4. Low-frequency component of the Moberg et al. (2005) multi-proxy reconstruction for Northern Hemisphere for AD 133–1925. PAGES2k reconstructed NH temperatures using different proxies that primarily shows high frequency resolution (PAGES 2k Consortium et al. 2019). Anomalies for both series are calculated in comparison to the 1961–1990 average.

Figure 5. Low-frequency bootstrapped summer temperature anomaly loess reconstruction from this study (blue) with standard error (gray) (top figure). Boxplots show estimates binned by century for all proxy samples (i.e., includes samples for the 100 simulations of the age-depth models). Anomalies are calculated in comparison to the 1961–1990 average. The temperature predictions using the median date from the age-depth model (i.e., does not include the 100 simulations) for each sample (not binned by century) are shown by circles; sizes of circles are inversely proportional to the sample-specific errors (size of circle ∝ 1/s1; see Simpson, 2007: 21 for calculation of s1). Anomaly estimates with larger circles are more precise. Histogram of proxy samples that includes the 100 simulations of the age-depth models through time in 100 year bins.

Figure 5. Low-frequency bootstrapped summer temperature anomaly loess reconstruction from this study (blue) with standard error (gray) (top figure). Boxplots show estimates binned by century for all proxy samples (i.e., includes samples for the 100 simulations of the age-depth models). Anomalies are calculated in comparison to the 1961–1990 average. The temperature predictions using the median date from the age-depth model (i.e., does not include the 100 simulations) for each sample (not binned by century) are shown by circles; sizes of circles are inversely proportional to the sample-specific errors (size of circle ∝ 1/s1; see Simpson, 2007: 21 for calculation of s1). Anomaly estimates with larger circles are more precise. Histogram of proxy samples that includes the 100 simulations of the age-depth models through time in 100 year bins.

Figure 6. Low-frequency July temperature anomaly for the SWUS from this study, the North American July temperature reconstruction (André E. Viau et al. 2006) from 3000 BC to AD 2000, the terrestrial composite for 30–60 °N (Kaufman et al. 2020) from 3300 BC to AD 1700 in 500-year bins, and summer temperature for western North American (Routson et al. 2021). All series standardized to a mean of 0 and an s of 1 over the period in this graph (Routson et al. 2021 was already standardized).

Figure 6. Low-frequency July temperature anomaly for the SWUS from this study, the North American July temperature reconstruction (André E. Viau et al. 2006) from 3000 BC to AD 2000, the terrestrial composite for 30–60 °N (Kaufman et al. 2020) from 3300 BC to AD 1700 in 500-year bins, and summer temperature for western North American (Routson et al. 2021). All series standardized to a mean of 0 and an s of 1 over the period in this graph (Routson et al. 2021 was already standardized).

Figure 7. Low-frequency July temperature anomaly for the SWUS from this study and the Northern Hemisphere low-frequency component (Moberg et al. 2005) from AD 133 to 1900. Pages2k reconstructed NH temperatures using different proxies that combine high and low frequency data (PAGES 2k Consortium et al. 2019). All series have been standardized to a mean of 0 and an s of 1 for this period.

Figure 7. Low-frequency July temperature anomaly for the SWUS from this study and the Northern Hemisphere low-frequency component (Moberg et al. 2005) from AD 133 to 1900. Pages2k reconstructed NH temperatures using different proxies that combine high and low frequency data (PAGES 2k Consortium et al. 2019). All series have been standardized to a mean of 0 and an s of 1 for this period.

Tables

Table 1. Metadata for the fossil pollen records (29 sites or 32 datasets) used in this study that have data after 3,550 BC. All data were downloaded from the Neotoma Paleoecology Database (J. W. Williams et al. 2018).
Site Name Site ID Dataset ID No. of Samples Elevation (m) Longitude Latitude
Alpine Pond 9814 14517 26 3028 -112.8245 37.63632
Alpine Pond 9814 14547 11 3028 -112.8245 37.63632
Bear Lake 10000 14951 12 2752 -112.1472 36.37112
Beef Pasture 246 250 90 3076 -108.1604 37.47310
Beef Pasture 246 25551 71 3076 -108.1604 37.47310
Cascade Fen [Engineer Mountain Bog] 339 347 63 2845 -107.8087 37.64764
Chihuahueños Bog 13680 20833 18 2862 -106.5105 36.04767
Columbine Ranch Fen 486 498 28 2793 -107.8157 37.60664
Como Lake 2957 3056 9 3651 -105.5142 37.56952
Crane Lake 504 517 7 2664 -112.1489 36.52986
Cumbres Bog 11617 17419 30 3115 -106.4505 37.02175
Fracas Lake 9997 14944 7 2530 -112.2386 36.63070
Hay Lake 982 1013 5 2764 -109.4250 34.00000
Head Lake 2955 3054 9 2295 -105.7406 37.71109
Hermit Lake 23930 41580 26 3701 -105.6317 38.08806
Hospital Flat Meadow 26738 46801 10 2875 -109.8774 32.66757
Hunters Lake 13486 20304 20 3576 -106.8437 37.61104
Jacob Lake 1127 1162 5 2326 -110.8333 34.33333
Leonora Curtin 14632 22948 12 1867 -106.1065 35.56948
Lowder Creek Bog 9830 14549 12 3186 -112.7922 37.67214
Molas Lake 1705 1761 25 3282 -107.6827 37.74759
Molas Pass Bog 1706 1762 15 3362 -107.6975 37.73778
Montezuma Well 1710 1766 31 1112 -111.7523 34.64920
Posy Lake 1905 1970 14 2693 -111.6960 37.93746
Potato Lake 1906 3560 14 2280 -111.3453 34.46222
Potato Lake 1906 3561 15 2280 -111.3453 34.46222
Purple Lake 13683 20838 39 3321 -111.5712 38.07438
San Agustin Plains 2260 3612 6 2115 -108.2500 33.86667
San Luis Lake 9878 14650 24 2295 -105.7247 37.67773
Stewart Bog 10191 15356 102 3083 -105.7220 35.83200
Stoneman Lake 10449 15967 17 2132 -111.5178 34.77887
Twin Lakes 2785 2880 54 3265 -108.1026 37.46906

Supplemental Materials

Supplemental Figures

Supplemental Figure S1. Squared residual length for the fossil assemblages (including the 100 age-depth model simulations) vs age. Red line marks the 95% limit of the calibration set residual lengths (very poorly fitted). We removed all samples that fell above the 95% line.

Supplemental Figure S1. Squared residual length for the fossil assemblages (including the 100 age-depth model simulations) vs age. Red line marks the 95% limit of the calibration set residual lengths (very poorly fitted). We removed all samples that fell above the 95% line.

Supplemental Figure S2. Summary diagram of the results of a MAT model applied to predict temperature from the modern pollen data set. The leave-one-out error measure (RMSEP) shows that 4 analogs provide the optimal choice (lowest RMSEP).

Supplemental Figure S2. Summary diagram of the results of a MAT model applied to predict temperature from the modern pollen data set. The leave-one-out error measure (RMSEP) shows that 4 analogs provide the optimal choice (lowest RMSEP).

Supplemental Figure S3. ime series plot of the minimum dissimilarity between each core (fossil) sample (including the 100 simulations of the age-depth models) and the modern pollen training set samples. The dotted, horizontal lines are drawn at various percentiles of the distribution of the pair-wise dissimilarities for the training set samples. We used the 5th percentile as the cutoff for good analogs. So, samples with distances greater than the 5th percentile (above the line) were removed.

Supplemental Figure S3. ime series plot of the minimum dissimilarity between each core (fossil) sample (including the 100 simulations of the age-depth models) and the modern pollen training set samples. The dotted, horizontal lines are drawn at various percentiles of the distribution of the pair-wise dissimilarities for the training set samples. We used the 5th percentile as the cutoff for good analogs. So, samples with distances greater than the 5th percentile (above the line) were removed.

Supplemental Figure S4. Linear regression and scatterplot between temperature anomaly predicted values and the bootstrapped temperature predictions. There is a strong linear relationship across their joint ranges (r2 = 0.96; p > F < .001).

Supplemental Figure S4. Linear regression and scatterplot between temperature anomaly predicted values and the bootstrapped temperature predictions. There is a strong linear relationship across their joint ranges (r2 = 0.96; p > F < .001).

Supplemental Figure S5. Histogram of fossil pollen samples using the median date in 100 year bins.

Supplemental Figure S5. Histogram of fossil pollen samples using the median date in 100 year bins.

Supplemental Figure S6. Low-frequency weighted bootstrapped temperature anomaly loess reconstruction from this study (blue) with standard error (gray). As in Figure 5 (main text) except here the loess curve (only) is weighted by the inverse of the sample-specific errors s1; see Simpson, 2007: 21 for calculation). CircleBubble size α 1/s1.

Supplemental Figure S6. Low-frequency weighted bootstrapped temperature anomaly loess reconstruction from this study (blue) with standard error (gray). As in Figure 5 (main text) except here the loess curve (only) is weighted by the inverse of the sample-specific errors s1; see Simpson, 2007: 21 for calculation). CircleBubble size α 1/s1.

Supplemental Tables

Supplemental Table S1. Summaries of MAT prediction output for the fossil pollen cores. An NA denotes that the site was removed after samples not passing quality tests.
Site Name Site ID Dataset ID Anomaly (°C) Anomaly Bootstrapped (°C) Mean boostrap RMSEP (°C) Mean boostrap s1
Alpine Pond 9814 14517 4.55 4.73 2.93 0.90
Alpine Pond 9814 14547 0.74 0.63 2.93 0.59
Bear Lake 10000 14951 9.59 7.61 3.22 2.30
Beef Pasture 246 250 11.77 10.93 4.66 2.28
Beef Pasture 246 25551 4.41 3.92 2.86 1.62
Cascade Fen [Engineer Mountain Bog] 339 347 4.86 4.50 3.09 1.09
Chihuahueños Bog 13680 20833 7.15 6.68 3.06 1.88
Columbine Ranch Fen 486 498 3.86 3.40 2.87 1.41
Como Lake 2957 3056 7.95 6.44 2.95 1.78
Crane Lake 504 517 5.00 2.63 2.81 1.84
Cumbres Bog 11617 17419 10.65 9.51 2.92 1.23
Hay Lake 982 1013 3.11 2.53 2.28 0.74
Head Lake 2955 3054 3.86 2.35 2.83 1.67
Hermit Lake 23930 41580 6.41 6.50 2.86 1.29
Hospital Flat Meadow 26738 46801 1.38 1.53 2.28 0.54
Hunters Lake 13486 20304 7.05 6.98 3.19 1.98
Jacob Lake 1127 1162 3.70 3.44 2.29 0.75
Leonora Curtin 14632 22948 4.14 3.30 2.75 1.48
Lowder Creek Bog 9830 14549 1.95 1.02 2.46 1.05
Molas Lake 1705 1761 5.29 4.61 2.55 0.98
Molas Pass Bog 1706 1762 2.11 1.64 2.93 1.81
Montezuma Well 1710 1766 3.35 2.82 2.79 1.32
Posy Lake 1905 1970 3.15 1.89 2.49 1.09
Potato Lake 1906 3560 3.18 2.95 2.95 1.47
Potato Lake 1906 3561 2.52 1.71 2.83 1.11
Purple Lake 13683 20838 2.38 1.53 2.49 0.62
San Luis Lake 9878 14650 5.66 4.96 2.90 1.84
Stewart Bog 10191 15356 9.60 9.84 4.18 1.26
Stoneman Lake 10449 15967 8.83 7.33 2.90 1.64
Twin Lakes 2785 2880 9.09 9.69 3.09 1.25
Fracas Lake 9997 14944 NA NA NA NA
San Agustin Plains 2260 3612 NA NA NA NA
Supplemental Table S2. Final reconstruction output from the loess model for 2950 BC to AD 1950.
Year Temperature Anomaly (°C) Confidence Interval
-2950 0.4045296 1.1618605
-2850 0.4414383 0.8050604
-2750 0.4718849 0.6307377
-2650 0.4881505 0.6074325
-2550 0.4882839 0.6430761
-2350 0.4696600 0.7031900
-2250 0.3847865 0.6944113
-2150 0.2431550 0.7145161
-2050 -0.0009257 0.6949540
-1950 -0.2118152 0.6957879
-1850 -0.3534270 0.6993471
-1750 -0.3573030 0.7294325
-1650 -0.1222834 0.7229518
-1550 0.1342259 0.7133825
-1450 0.3711475 0.7102845
-1350 0.4147778 0.6993040
-1250 0.1310486 0.6837305
-1150 -0.3198836 0.6631538
-1050 -0.6173994 0.6619437
-950 -0.5141464 0.6750019
-850 -0.0207968 0.6890323
-750 0.4170984 0.7027943
-650 0.5032538 0.7072723
-550 0.3552907 0.7093661
-450 0.2172818 0.6957468
-350 0.2512441 0.6866448
-150 0.3198603 0.6869330
-50 0.3804943 0.7066400
50 0.2724975 0.6918105
150 -0.0347638 0.7089141
250 -0.4984228 0.6795270
350 -0.7884595 0.6694945
450 -0.6936688 0.6897010
550 -0.3664989 0.6932318
650 -0.1413861 0.7281681
750 -0.1150580 0.7154483
850 -0.2414175 0.7001981
950 -0.5253352 0.6678101
1050 -0.1924888 0.6897357
1150 0.2362539 0.7070884
1250 0.0836372 0.7147397
1350 0.0733919 0.6726624
1450 -0.0210632 0.6557371
1550 0.0750696 0.6902302
1650 0.5968028 0.7251663
1750 0.8151299 0.7139425
1850 0.9663412 0.6594453
1950 1.4964440 0.8462535